Integrative Gene-Enhancer Association Scoring

Gene–enhancer associations were generated based on five methods: eQTLs (45 (link)), eRNA co-expression (22 (link)), TF co-expression, capture Hi-C (CHi-C) (30 (link)) and gene target distance, all of which are described in the Supplementary Methods. Subsequently, a score S_GE was calculated for each gene–enhancer link, to estimate the strength of such connection. S_GE is defined as:
$S_{GE}=-{\text{Log}}_{10}\left(p_g\right)+S_C+c\cdot f$
where p_g is a combined P-value for eQTLs, eRNA co-expression and TF co-expression, computed by Fisher’s combined probability test via a χ² test statistic (46 ). The second term (S_C) represents the CHi-C score as provided by the source, constituting the logarithm of the ratio of observed to expected read counts (30 (link)). The third term is related to enhancer–gene distance, where c is a normalization score based on the average score from the first two terms across all gene–enhancer connections. To compute f we draw a gene–enhancer distance distribution (Supplementary Figure S8), and obtain f as the fraction of enhancers in the distance bin of the specific gene–enhancer pair. Gene–enhancer distances are computed between a gene’s TSS and the mid-point of an enhancer, and the distribution employed for the purpose of computing f excludes values from the CHi-C method, which lacks information in the crucial range of 0–20 kb.
Our method for computing S_GE is on the whole unbiased, and minimally involves arbitrary weighting factors. The three scores for eQTLs, eRNA co-expression and CHi-C are based on the reported summary statistics and the significance thresholds used in the original studies. For TF co-expression we computed P-values as shown in the Supplementary Methods (‘Transcription factor co-expression analysis’ paragraph). When possible, P-values were combined in a meta-analytic fashion, using the widely utilized Fisher’s combined probability test.